A method and system for real-time assessment of quality of cardiopulmonary resuscitation based on video analysis

Abstract

The application discloses a kind of based on video analysis cardiopulmonary resuscitation quality real-time evaluation method and system, it is related to video processing technical field, including real-time analysis to cardiopulmonary resuscitation video and extract operation key point, and the validity of data is judged by key point extraction evaluation and continuity detection.When key point extraction is invalid, based on key cause distinguishes operation occlusion and trajectory tracking anomaly, and respectively executes key point prediction or trajectory tracking optimization adjustment, realizes key point completion and trajectory repair, generates real-time optimization key point and optimization trajectory sequence, and carries out cardiopulmonary resuscitation operation quality evaluation and information feedback. Whether the result is improved by key point extraction to judge whether to carry out trajectory optimization secondary adjustment, realizes iterative optimization.The application solves the problem that cardiopulmonary resuscitation operation key point continuity motion trajectory detection exists intermittency interruption, leading to the problem of poor cardiopulmonary resuscitation quality evaluation precision.

Description

Technical Field

[0001] This invention relates to the field of video processing technology, specifically to a method and system for real-time assessment of cardiopulmonary resuscitation quality based on video analysis. Background Technology

[0002] Existing video-based human motion analysis technologies typically extract key points of the human skeleton through pose estimation algorithms and combine them with joint angles, spatial positions, or posture features for motion recognition or functional assessment. They mainly extract sequences of human key points from videos and model these key points through feature engineering or deep learning models to output motion categories or posture evaluation results.

[0003] For example, Chinese invention patent CN110222665B discloses a method for human motion recognition in surveillance based on deep learning and pose estimation. The method includes: constructing a multi-stream motion recognition model based on relative joint feature representations; preprocessing and converting the relative joint feature representations of human skeletal motion data; inputting the converted relative joint feature representations into the multi-stream motion recognition model for training and evaluation, selecting the optimal model with the highest convergence recognition rate after multiple iterations; acquiring surveillance segments in real-time video scenes, using pose estimation algorithms to obtain the skeletal motion sequence of the human body in the surveillance segment, and preprocessing it; converting the feature representation of the preprocessed skeletal motion sequence; using the optimal model to identify human motions in the preprocessed and feature-converted skeletal motion sequence, obtaining motion classification results; comparing the identified classification results with preset dangerous motion categories, and returning the comparison results to the monitoring staff.

[0004] For example, Chinese invention patent CN110969114B discloses a human motion function detection system, detection method, and detection instrument, which includes: a data acquisition module that uses a depth camera to acquire video of human movement; a skeletal node position acquisition module for establishing a simplified human skeletal model for analysis and acquiring spatial coordinate data of the human skeleton; a data calculation module for obtaining spatial data between joints based on the depth position of each coordinate point in the skeletal data; a posture database matching module for matching the spatial data with a posture database template obtained through machine learning from a large number of data samples; and a limb movement recognition module for performing limb recognition.

[0005] However, in emergency department cardiopulmonary resuscitation (CPR) scenarios, CPR is a continuous, periodic action with strict rhythmic requirements. Its quality assessment relies on dynamic parameters such as compression frequency, compression depth, and rebound. Existing technologies primarily focus on action category recognition or static posture matching, lacking the ability to temporally model the continuous motion trajectory of key points. In practical applications, due to rescuer obstruction, posture changes, and limitations of video acquisition angles, key point detection results are prone to discontinuity, drift, or missing values, leading to discontinuous or abnormally abrupt key point trajectories. This interferes with the calculation of dynamic parameters based on key points, resulting in inaccurate compression cycle recognition and large parameter fluctuations, thus affecting the reliability of CPR quality assessment results. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method and system for real-time assessment of cardiopulmonary resuscitation quality based on video analysis.

[0007] To achieve the above objectives, the technical solution of the present invention is as follows:

[0008] In a first aspect, this invention discloses a real-time assessment method for cardiopulmonary resuscitation (CPR) quality based on video analysis, comprising the following steps: performing real-time video analysis and extracting key operational points from CPR videos to obtain each key operational point, and performing effective assessment and continuity detection based on each key operational point to obtain key point extraction assessment results; performing CPR quality assessment to obtain CPR operation quality values; when the key point extraction assessment result is invalid, marking the CPR operation quality value as an invalid CPR operation value, and retrieving the key causes of the key point extraction assessment results; otherwise, not performing the marking; when the key cause is operation occlusion, performing key point prediction to obtain predicted filling. Key points: When the key cause is abnormal trajectory tracking, trajectory tracking optimization and adjustment are performed to obtain optimized key points; based on the invalid value of cardiopulmonary resuscitation operation, real-time updates and trajectory optimization adjustments are performed to obtain real-time optimized key points and their corresponding optimized trajectory sequences; based on the analysis of predicted key points or tracked optimized key points, the quality value of cardiopulmonary resuscitation operation is obtained, and corresponding information reminders are issued; based on the analysis of real-time optimized key points and their corresponding optimized trajectory sequences, the key point extraction and improvement results are obtained. If the key point extraction and improvement results are unsuccessful, a second trajectory optimization adjustment is performed until the key point extraction and improvement results are successful; otherwise, no processing is performed.

[0009] Secondly, this invention discloses a real-time cardiopulmonary resuscitation (CPR) quality assessment system based on video analysis, comprising the following modules: a key point extraction module, used for real-time video analysis and extraction of key points from CPR videos to obtain key points for each operation, and for effective evaluation and continuity detection based on each key point to obtain key point extraction assessment results; an initial assessment module, used for performing CPR quality assessment to obtain CPR operation quality values, and when the key point extraction assessment result is invalid, marking the CPR operation quality value as an invalid CPR operation value and retrieving the key causes of the key point extraction assessment result; otherwise, the marking is not performed; and a key cause determination module, used for performing key point prediction when the key cause is operation occlusion to obtain predicted filling key points. Key points; when the key cause is trajectory tracking anomaly, trajectory tracking optimization and adjustment are performed to obtain tracking optimization key points; the trajectory tracking adjustment module is used to perform real-time update acquisition and trajectory optimization adjustment of operation key points based on the invalid value of cardiopulmonary resuscitation operation, to obtain real-time optimized key points and their corresponding optimized trajectory sequences; the operation evaluation and reminder module is used to obtain the cardiopulmonary resuscitation operation quality value based on the prediction filling key points or the analysis of tracking optimization key points, and issue corresponding information reminders; the secondary adjustment module is used to obtain the key point extraction improvement result based on the analysis of real-time optimized key points and their corresponding optimized trajectory sequences. If the key point extraction improvement result is an improvement failure, the trajectory optimization secondary adjustment is performed until the key point extraction improvement result is an improvement success, otherwise no processing is performed.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0011] 1. This invention performs real-time video analysis and extracts and evaluates key points of cardiopulmonary resuscitation (CPR) videos. When the key point extraction evaluation result is invalid, it retrieves the key causes and distinguishes between operation occlusion and trajectory tracking anomalies. This enables adaptive identification and correction of missing, drifting, and discontinuous key points of the executor in CPR video quality analysis, effectively solving the problem of intermittent interruptions in the continuous motion trajectory detection of key points in CPR operations in existing technologies, which leads to poor accuracy in CPR quality assessment.

[0012] 2. This invention performs key point prediction and generates predicted filling key points when the key cause is determined to be operation occlusion. Based on historical key point trajectory data, it performs periodic feature detection and phase trend prediction or motion trend continuation, thereby realizing continuous completion of key point information and trajectory recovery under occlusion.

[0013] 3. By performing trajectory tracking optimization and adjustment when the key cause is determined to be an abnormality in trajectory tracking, and by smoothing the time series, correcting abnormal jump points, and replacing interpolation coordinates, the corrected key point trajectory is generated and the periodic trajectory is reconstructed, thereby achieving enhanced stability of key point motion trajectory and suppression of abnormal jumps. Attached Figure Description

[0014] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals refer to the same parts. Wherein:

[0015] Figure 1 This is an overall flowchart of the real-time cardiopulmonary resuscitation quality assessment method of the present invention;

[0016] Figure 2 This is a flowchart illustrating the trajectory tracking optimization and adjustment process of the present invention.

[0017] Figure 3 This is a system architecture diagram of the present invention. Detailed Implementation

[0018] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0019] This invention targets scenarios such as emergency training, clinical assistance, and routine CPR instruction. It extracts key operational points from video data of the CPR process in real time and constructs the motion trajectories of these key points, enabling dynamic assessment of core quality parameters such as compression frequency, compression depth, and rebound. Existing video-based human motion analysis technologies mostly focus on posture recognition or action classification, lacking the ability to model the temporal sequence of continuous periodic movements, making it difficult to accurately extract dynamic quality indicators during CPR. Furthermore, in practical applications, due to rescuer obstruction and detection noise, key point data suffers from discontinuities and jumps, leading to unstable trajectory structures and affecting the accuracy and reliability of the assessment results.

[0020] To address the aforementioned issues, this invention proposes a comprehensive evaluation method combining keypoint validity detection, cause determination, trajectory restoration, and temporal optimization. By constructing a keypoint confidence determination and continuity detection mechanism, the effectiveness of keypoint extraction is assessed. Furthermore, continuous invalid frame analysis distinguishes between occlusion and trajectory anomalies. For occlusion, a periodic prediction and trend prediction mechanism based on historical trajectories is introduced for keypoint compensation. For trajectory anomalies, methods such as smoothing, anomaly detection, neighborhood interpolation, and periodic reconstruction are employed to restore the keypoint trajectory structure. Based on this, an adaptive adjustment mechanism based on quality feedback is introduced to optimize the temporal resolution of the keypoint trajectory by dynamically adjusting the detection frame extraction interval and time resampling step size. When the optimization effect is insufficient, secondary optimization is performed through secondary backtracking duration expansion and neighborhood interpolation range enhancement to further improve trajectory stability. This invention can effectively restore the temporal structure of key points in complex environments, improve keypoint continuity and stability, thereby enhancing the accuracy and robustness of cardiopulmonary resuscitation quality assessment.

[0021] This invention includes the following steps: performing real-time video analysis and key point extraction on cardiopulmonary resuscitation (CPR) videos to obtain each key point, and conducting effective evaluation and continuity detection based on each key point to obtain key point extraction evaluation results; performing CPR quality assessment to obtain CPR operation quality values; when the key point extraction evaluation result is invalid, marking the CPR operation quality value as an invalid value, and retrieving the key cause of the key point extraction evaluation result; otherwise, the marking is not performed; when the key cause is operation occlusion, performing key point prediction to obtain predicted and filled key points; when the key cause is trajectory tracking anomalies... Normally, trajectory tracking optimization and adjustment are performed to obtain tracking optimization key points; based on the invalid value of cardiopulmonary resuscitation operation, real-time updates and trajectory optimization adjustments are performed to obtain real-time optimized key points and their corresponding optimized trajectory sequences; based on the analysis of predicted key points or tracking optimization key points, the quality value of cardiopulmonary resuscitation operation is obtained, and corresponding information reminders are issued; based on the analysis of real-time optimized key points and their corresponding optimized trajectory sequences, the key point extraction improvement result is obtained. If the key point extraction improvement result is an improvement failure, a second trajectory optimization adjustment is performed until the key point extraction improvement result is an improvement success; otherwise, no processing is performed.

[0022] In this embodiment, as Figure 1 As shown, Figure 1The following is an overall flowchart of the real-time cardiopulmonary resuscitation (CPR) quality assessment method of the present invention. The method proceeds as follows: Real-time video analysis and extraction of key operational points are performed on the CPR video, and key point extraction assessment and continuity detection are conducted for each key operational point. Then, based on the key point extraction assessment results, branch judgment is made. When the assessment result indicates valid extraction, the process directly proceeds to the CPR operation quality value calculation stage. When the assessment result indicates invalid extraction, key cause analysis is further performed, and the type of key cause is determined. When the key cause type is operation occlusion, key point prediction is performed to obtain predicted filling key points. When the key cause type is trajectory tracking anomaly, trajectory tracking optimization adjustment is performed to obtain tracking optimization key points. After obtaining the predicted filling key points or tracking optimization key points, real-time optimized key points and their corresponding optimized trajectory sequences are generated, and CPR operation quality value is calculated and information is output based on the results. Subsequently, the key point extraction improvement results are analyzed. When the improvement result indicates improvement failure, secondary trajectory optimization adjustment is performed, and the process returns to the optimization process to regenerate real-time optimized key points and optimized trajectory sequences. When the improvement result indicates improvement success, the process ends.

[0023] It should be noted that the periodicity characteristic determination result refers to whether the data exhibits periodic changes. This data can be historical data or real-time collected data.

[0024] When the key point extraction assessment result is valid, cardiopulmonary resuscitation (CPR) quality assessment is performed directly, and the obtained CPR operation quality value is the same as that based on prediction parameters and methods.

[0025] Key points of operation are the critical joints on the human body involved in the compression motion during CPR, specifically including key points such as the wrist, elbow, and shoulder. Human pose recognition algorithms (such as OpenPose) can detect these key points in video frames and output two-dimensional coordinates. Predictive key point filling refers to key points predicted from historical trajectories even in occluded situations. Tracking-optimized key points refer to key points obtained through trajectory repair and optimization. Real-time optimized key points refer to key points after time resampling and parameter adjustment.

[0026] Furthermore, the key point extraction evaluation results are obtained through the following method: Each operational key point within a preset continuous detection frame is acquired, and key points of the same type are aligned in chronological order to obtain a temporal sequence of operational key points; frame-by-frame confidence testing is performed on the operational key points in each detection frame to obtain the confidence level of each operational key point within each detection frame; the confidence level of each operational key point within each detection frame is compared frame-by-frame with a preset confidence threshold to obtain the validity judgment result of each detection frame. If the confidence level of each operational key point within a detection frame is above the confidence threshold, the validity judgment result of that detection frame is considered reliable; otherwise, the validity judgment result of that detection frame is considered unreliable. The validity judgment results of a preset number of consecutive detection frames are extracted. If the validity judgment results of the preset number of consecutive detection frames are all reliable, the continuity detection result is considered continuous; otherwise, the continuity detection result is considered interrupted. When the validity judgment result is unreliable and / or the continuity detection result is considered interrupted, the key point extraction evaluation result is considered invalid; otherwise, the key point extraction evaluation result is considered valid.

[0027] In this embodiment, the timing sequence of each operation key point is obtained by arranging the operation key points representing the same operation key point in each detection frame in chronological order.

[0028] Frame-by-frame confidence testing is performed on the key points of operation in each detection frame to obtain the confidence level of each key point within each detection frame. Specifically, each frame image is input into a human pose recognition model, and human features are extracted using a convolutional neural network. For each predefined key point (such as wrist, elbow, etc.), the model outputs a corresponding two-dimensional probability heatmap. The maximum response value method is used to determine the location of the key point, and the maximum value of the heatmap corresponding to that location is taken as the confidence level of that key point. The above steps are repeated for all key points in each frame to obtain the set of confidence levels for all key points of operation within that frame.

[0029] By performing frame-by-frame confidence testing on key operational points in each detection frame, and combining this with validity analysis of consecutive detection frames, a comprehensive evaluation of the reliability and continuity of key point extraction results can be achieved. Analyzing key point confidence allows for a quantitative assessment of the reliability of each key point's detection result, thus determining whether the key point's location in the current frame is accurate and reliable. Since human pose estimation models are prone to false detections or shifts in complex environments (such as occlusion, lighting changes, and pose changes), relying solely on the key point location itself cannot guarantee data validity. Confidence threshold filtering effectively eliminates low-confidence key points, preventing unreasonable data from participating in subsequent trajectory analysis, thereby ensuring the rationality of the key point extraction results. Statistical analysis of the validity of consecutive detection frames further determines the continuity of key point data over time. Because cardiopulmonary resuscitation (CPR) is a periodic and continuous action, frequent interruptions or absences of key points in the time series indicate instability in the key point extraction process, directly affecting the accuracy of subsequent calculations of dynamic parameters such as compression frequency and depth. Therefore, by determining the continuous valid frames, continuity defects in the key point extraction process can be identified, thus providing a basis for subsequent trajectory repair and optimization.

[0030] Furthermore, the key causes of the key point extraction evaluation results are retrieved. The specific method is as follows: trace back from the real-time frame, count the number of consecutive detection frames that are invalid in the key point extraction evaluation results, and record them as the number of consecutive invalid frames; compare the number of consecutive invalid frames with a preset occlusion judgment frame number threshold. When the number of consecutive invalid frames is above the occlusion judgment frame number threshold, the key cause of the key point extraction evaluation results is determined to be operation occlusion; when the number of consecutive invalid frames is less than the occlusion judgment frame number threshold, the key cause of the key point extraction evaluation results is determined to be trajectory tracking anomaly.

[0031] In this embodiment, the number of consecutive detection frames deemed invalid by keypoint extraction is statistically analyzed. Based on a comparison of this number with a preset threshold, the cause of keypoint anomalies is automatically determined, thus distinguishing between operational occlusion and trajectory tracking anomalies. In actual cardiopulmonary resuscitation (CPR), different types of anomalies exhibit different characteristics over time. When the rescuer's body obstructs the view, keypoints are typically missing or have reduced confidence in multiple consecutive detection frames, resulting in a prolonged period of consecutive invalid frames. Trajectory tracking anomalies, on the other hand, are usually caused by detection errors or local jitter, with a smaller impact range, manifesting as short-term, scattered invalid frames or local jumps, without causing prolonged continuous interruptions. Therefore, by analyzing the number of consecutive invalid frames, occlusion-related problems and trajectory anomaly problems can be effectively distinguished.

[0032] Since occlusion problems and trajectory tracking anomalies are essentially two different types of problems, their corresponding repair methods also differ significantly: occlusion problems require predictive compensation based on historical trajectories, while trajectory anomalies require repair through smoothing, interpolation, and trajectory reconstruction. If the causes of anomalies are not differentiated and treated uniformly, it can easily lead to mismatched repair strategies, thus affecting the recovery of key point trajectories. Therefore, this step categorizes and determines the causes of anomalies, enabling targeted selection of subsequent processing strategies, improving the accuracy and effectiveness of key point repair, and ultimately enhancing the overall precision and stability of CPR quality assessment.

[0033] Furthermore, keypoint prediction is performed to obtain predicted and filled keypoints. Specifically, the key causes of the keypoint extraction evaluation results are retrieved, specifically the detection frames corresponding to operation occlusion, and these are marked as operation occlusion detection frames. The first occurrence time corresponding to this operation occlusion detection frame is recorded as the operation occlusion start time. Starting from the operation occlusion start time and with a preset first backtracking processing duration as the backtracking execution duration, data backtracking processing is performed on the operation keypoints to obtain historical keypoint trajectory data. This historical keypoint trajectory data includes the historical position sequence of each operation keypoint. Based on the historical keypoint trajectory data, periodic feature detection is performed to obtain the weekly... The periodicity feature determination result is as follows: When the periodicity feature determination result indicates the presence of periodicity, phase trend prediction is performed based on historical keypoint trajectory data to obtain the predicted keypoint position values ​​for each frame within the operation occlusion detection frame, and these values ​​are used as periodic prediction filling keypoints; when the periodicity feature determination result indicates the absence of periodicity, the movement direction and speed of the operation keypoints in the detection frame are analyzed based on historical keypoint trajectory data, and the operation keypoint positions for each detection frame within the operation occlusion detection frame are continued along the movement trend to obtain trend prediction filling keypoints; the periodic prediction filling keypoints and / or trend prediction filling keypoints are used as predicted filling keypoints.

[0034] In this embodiment, it should be noted that the predicted value of the key point location represents the predicted coordinates of the key point location.

[0035] Historical key point trajectory data refers to the data obtained by tracing back from the start time of operation occlusion to the first backtracking processing time, obtaining historical backtracking detection frames, sorting the historical backtracking detection frames by time, extracting the data of each corresponding operation key point according to the sorting order, and arranging them to obtain the historical data of each operation key point, i.e., historical key point trajectory data.

[0036] Periodic feature detection is performed based on historical keypoint trajectory data to obtain periodic feature determination results. Specifically, the displacement data of each operational keypoint in the vertical direction (Y-axis) is extracted from the historical keypoint trajectory data to form a time series. This series is then processed to remove the mean. The autocorrelation coefficient is calculated using the series itself and its delayed sequence. The method for performing periodic feature detection on the corrected keypoint trajectory is the same as described above, involving the calculation of the autocorrelation coefficient. The specific formula for calculating the autocorrelation coefficient is as follows:

[0037] ;

[0038] In the formula, Represents the autocorrelation coefficient. Indicates the number of delayed frames. Indicates the first Vertical displacement values ​​of key points in frame operation Indicates the number of the detection frame. , This represents the total number of frames in the historical key point trajectory data. Representing time series The arithmetic mean, This represents the time series of displacement data of each key operational point in the vertical direction.

[0039] Peak extraction is performed on the time series to search for the delay point corresponding to the first significant peak, and the delay point is compared with the preset periodic correlation threshold. If the delay point is greater than the periodic correlation threshold, the periodic feature determination result is that there is a periodic feature, and the delay point is the pressing period step.

[0040] The predicted keypoint positions for each frame within the operation occlusion detection frame are obtained as follows: Based on historical keypoint trajectory data, the average amplitude (half the compression depth) and average equilibrium position (i.e., the coordinates of the compression median line during CPR) of the most recent three complete cycles are obtained through extreme point detection. Using the data of the frame preceding the start of the operation occlusion as a reference, the phase of that point in the standard cycle waveform is calculated. The predicted keypoint position value is then calculated for each frame within the operation occlusion detection frame, using the following formula:

[0041] ;

[0042] In the formula, Indicates the first Predicted location coordinates of intra-frame key points in each operation occlusion detection frame. This indicates the number of the operation occlusion detection frame. , This indicates the operation of the occlusion detection frame. Indicates the equilibrium position of the historical trajectory. This represents the average amplitude of the historical trajectory. Indicates the period step size. This indicates the initial phase corresponding to the start time of the occlusion operation. It represents pi (π).

[0043] The calculated predicted position coordinates are sequentially filled into the corresponding detection frames to form periodic prediction filling key points.

[0044] To obtain trend predictions and fill in key points, the specific method is as follows:

[0045] ;

[0046] In the formula, Indicates the first Predicted coordinates of key points in each operation occlusion detection frame. This indicates the last known coordinates of the key points at the start of the occlusion operation. Indicates the instantaneous velocity at which the occlusion occurs. This indicates the number of the operation occlusion detection frame. , This indicates the total number of operation occlusion detection frames. Indicates the sampling time interval of the video. It represents the instantaneous acceleration of motion when the blockage occurs.

[0047] By performing periodic feature analysis and aperiodic trend analysis on historical keypoint trajectory data, the system can predict and complete keypoint positions even under occlusion conditions. When periodic features exist, a periodic phase continuation method is used to ensure the prediction results remain consistent with the original compression rhythm; when no periodic features exist, a trend continuation method is used to ensure the prediction results match the actual movement direction changes. Through different prediction mechanisms, the trajectory of keypoints within the detected occlusion frame remains continuous, thus avoiding interference from missing keypoints caused by occlusion in the calculation of CPR operation quality values, improving the stability and accuracy of the assessment. Simultaneously, the system's robustness in complex occlusion environments is enhanced, enabling stable output of CPR operation quality assessments even under non-ideal video acquisition conditions.

[0048] Further, trajectory tracking optimization and adjustment are performed to obtain tracking optimization key points. Specifically, the time sequence of each operational key point is smoothed to obtain a smoothed key point trajectory; anomaly detection and correction are performed on the smoothed key point trajectory, identifying operational key points whose positional abrupt changes exceed a preset abrupt change threshold and designating them as abrupt jump points; the coordinates of the abrupt jump points are replaced with interpolated coordinates calculated based on adjacent valid detection frames to obtain the corrected operational key point motion trajectory, which is recorded as the corrected key point trajectory; periodic feature detection is performed on the corrected key point trajectory to obtain the periodic feature determination result; if the periodic feature determination result... If periodic features are present, a preset number of reference trajectory segments are selected from the corrected keypoint trajectory, and periodic trajectory reconstruction is performed based on the reference trajectory segments to obtain the periodic reconstructed trajectory. If the periodic feature determination result is that no periodic features exist, an early warning is issued. The reference trajectory segments are specifically continuous trajectory segments with a period length consistency higher than the period length consistency threshold, i.e., continuous trajectory segments with an autocorrelation coefficient greater than the preset autocorrelation coefficient threshold. The corrected keypoint trajectory is corrected based on the periodic reconstructed trajectory to obtain the keypoint reconstructed trajectory. Each keypoint in the keypoint reconstructed trajectory is used as the tracking optimization keypoint, and its corresponding phase data is used as the tracking optimization keypoint data.

[0049] In this embodiment, as Figure 2 As shown, Figure 2 The flowchart for trajectory tracking optimization and adjustment of this invention is as follows: A smooth keypoint trajectory is acquired, and abnormal jump point detection is performed based on this trajectory. Interpolation coordinate replacement processing is applied to the detected abnormal jump points to generate a corrected keypoint trajectory. Subsequently, periodic feature detection is performed on the corrected keypoint trajectory. When the detection result indicates the presence of periodic features, a reference trajectory segment is extracted and periodic trajectory reconstruction is performed to generate a reconstructed keypoint trajectory. When the detection result indicates the absence of periodic features, the corrected keypoint trajectory is directly used as the reconstruction basis. Finally, the optimized keypoints are output, achieving optimized output of the keypoint trajectory.

[0050] The smooth key point trajectory is obtained by performing a moving average filter on the coordination sequence of each operation key point. That is, the coordinate values ​​within the window are taken as the window size with the preset smoothing window length, and the window is smoothed successively along the time direction to obtain the smooth key point waveform trajectory after filtering out high-frequency noise.

[0051] The specific method for obtaining abnormal transition points is as follows: calculate the Euclidean distance between the same operational key points in two adjacent frames of smooth key points frame by frame (that is, calculate the Euclidean distance between the coordinates of the key point in the next frame and the coordinates of the key point in the previous frame). If the Euclidean distance exceeds the preset abrupt change amplitude threshold, then the key point in the current frame is marked as an abnormal transition point.

[0052] The interpolated coordinates calculated based on adjacent valid detection frames are obtained as follows: For the detection frame marked as an abnormal jump point, in the time sequence of its corresponding operation key points, the nearest valid detection frame is searched forward along the time sequence, and the nearest valid detection frame is searched backward along the time sequence. The above two detection frames are respectively taken as the forward valid detection frame and the backward valid detection frame; the coordinate values ​​of the corresponding operation key points in the forward valid detection frame and the backward valid detection frame are extracted, and based on the time position relationship between the two valid detection frames, linear interpolation calculation is performed on the key point coordinates of the detection frame to which the abnormal jump point belongs to obtain the replacement coordinates of the abnormal jump point.

[0053] ;

[0054] ;

[0055] In the formula, The x-coordinate represents the replacement coordinates of the abnormal frame. The ordinate represents the replacement coordinates of the abnormal frame. This represents the x-coordinate of the valid forward detection frame corresponding to the anomalous transition point. This represents the ordinate of the forward valid detection frame corresponding to the anomalous transition point. Indicates the frame to which the abnormal transition point belongs. This represents the valid forward detection frame corresponding to the abnormal transition point. This represents the valid backward detection frame corresponding to the abnormal transition point. This represents the x-coordinate of the valid backward detection frame corresponding to the anomalous transition point. This represents the ordinate of the backward valid detection frame corresponding to the abnormal transition point.

[0056] The corrected motion trajectory of the key points is obtained and recorded as the corrected key point trajectory. The specific method is as follows: after replacing the coordinates of the abnormal jump points with the interpolated coordinates calculated based on the adjacent valid detection frames, the temporal sequence of the replaced key points is locally smoothed. Specifically, with the replacement point as the center, a time window of a preset length is selected, and the key point coordinate sequence within the window is processed by a sliding weighted average to eliminate the possible local discontinuities between the interpolated points and the original valid detection frames. Finally, the key point coordinates of each detection frame after processing are connected in chronological order to obtain the motion trajectory of the key points with enhanced continuity and smooth changes, which is recorded as the corrected key point trajectory.

[0057] The periodic reconstructed trajectory is obtained by performing uniform length normalization on each reference trajectory segment. Specifically, according to the number of press cycle frames within each reference trajectory segment, each reference trajectory segment is resampled to the same standard frame length through linear interpolation, thus obtaining a sequence of reference trajectory segments with consistent length. Subsequently, the keypoint coordinates at the same time index position of each reference trajectory segment are statistically averaged point by point to obtain a standard periodic keypoint sequence. Based on this standard periodic keypoint sequence, according to the period division results in the original corrected keypoint trajectory, the standard periodic keypoint sequence is repeatedly extended according to the period length, and replaced frame by frame with the corresponding period intervals in the corrected keypoint trajectory, so that the keypoint trajectories of each period segment are consistent in shape. Finally, a keypoint motion trajectory with a unified periodic structure and consistent rhythm is obtained, which is recorded as the periodic reconstructed trajectory.

[0058] The keypoint reconstruction trajectory is obtained as follows: Using the periodic reconstruction trajectory as a reference trajectory, the keypoints at corresponding time positions in the corrected keypoint trajectory are calculated frame-by-frame to obtain the keypoint offset for each detection frame. Based on this keypoint offset, a weighted adjustment process is performed on the keypoint coordinates in the corrected keypoint trajectory. Specifically, according to a preset adjustment ratio, the corrected keypoint coordinates are linearly updated along the offset direction towards the corresponding positions in the periodic reconstruction trajectory, thereby gradually reducing the deviation between the keypoints and the periodic reconstruction trajectory. During the adjustment process, the displacement changes of keypoints between adjacent detection frames are constrained. When the displacement changes between adjacent frames after adjustment exceed a preset continuity threshold, local smoothing is performed on the keypoint to avoid generating new abnormal jump points. Finally, the adjusted keypoints of each detection frame are connected in chronological order to obtain a keypoint motion trajectory that combines periodic consistency and trajectory continuity, which is recorded as the keypoint reconstruction trajectory.

[0059] By detecting anomalies in the smooth keypoint trajectory, abrupt positional changes caused by video noise, residual occlusion, or tracking errors can be identified. Interpolation based on adjacent valid detection frames effectively eliminates unrealistic jumps in the trajectory, ensuring the continuity and authenticity of the basic data. Periodic feature detection and selection of reference trajectory segments with high consistency in cycle length enable periodic trajectory reconstruction, making the trajectory shape between different pressing cycles more consistent, thus solving the problem of cycle instability caused by operational rhythm fluctuations or detection errors. The reconstructed trajectory is then used to correct the modified keypoint trajectory, making the keypoints approach the standard periodic trajectory frame by frame, achieving overall trajectory consistency optimization. The resulting optimized tracking keypoints not only possess higher spatial continuity and temporal stability, but their corresponding phase data also accurately reflects the positional state of the pressing action within the cycle, thereby improving the usability and reliability of the keypoint data analysis.

[0060] By employing self-repair and self-optimization of keypoint trajectories in video environments, the system can still output stable and reliable CPR keypoint data even in the presence of occlusion interference, tracking anomalies, or data fluctuations. The introduction of anomaly jump point analysis and correction mechanisms solves the problem of trajectory breakage caused by abrupt changes in keypoints in traditional video analysis. Periodic trajectory reconstruction gives keypoint trajectories a unified rhythmic structure in the time dimension, enhancing the accuracy of compression cycle recognition. Through keypoint reconstruction and phase data output, keypoints not only possess spatial location significance but also clear periodic stage identifiers, enabling subsequent CPR quality value calculations to be based on more complete, continuous, and rhythmically informative keypoint trajectories, avoiding evaluation bias caused by data anomalies or rhythm disturbances.

[0061] Furthermore, the real-time optimization key points and their corresponding optimization trajectory sequences are obtained. Specifically, the method is as follows: A difference analysis is performed based on the invalid values ​​of CPR operations and a preset CPR operation quality threshold to obtain the operation quality gap value; a preset operation quality gap threshold is obtained and compared with the operation quality gap value. If the operation quality gap value is below the threshold, the detection interval adjustment value of the detection frame is obtained based on the invalid CPR operation value matching; if the operation quality gap value is greater than the threshold, the detection interval adjustment value of the detection frame is obtained based on the operation quality gap value matching. The interpolation resampling step size is adjusted; the detection frame extraction interval is shortened based on the detection interval adjustment value by subtracting the interpolation resampling step size adjustment value from the current detection frame extraction interval to obtain the adjusted detection frame extraction interval; the interpolation resampling step size is reduced based on the interpolation resampling step size adjustment value by subtracting the interpolation resampling step size adjustment value from the current interpolation resampling step size to obtain the adjusted interpolation resampling step size; based on the adjusted detection frame extraction interval or interpolation resampling step size, time resampling interpolation is performed on the key point trajectory sequence corresponding to the key point of the operation to obtain the real-time optimized key point and its corresponding optimized trajectory sequence.

[0062] In this embodiment, the operation quality gap value is obtained by subtracting the cardiopulmonary resuscitation operation quality threshold from the cardiopulmonary resuscitation operation invalid value and taking the absolute value to obtain the cardiopulmonary resuscitation operation quality difference value. The operation quality gap value is obtained by dividing the cardiopulmonary resuscitation operation quality difference value by the cardiopulmonary resuscitation operation quality threshold value.

[0063] Based on the matching of invalid values ​​of cardiopulmonary resuscitation (CPR) operations, the detection interval adjustment value of the detection frame is obtained. The specific method is as follows: obtain the preset invalid value ranges of each CPR operation and the historical value of the detection interval adjustment corresponding to each invalid value range in the database, and match them with the invalid values ​​of CPR operations. If the invalid value of CPR operation is within a certain preset invalid value range of CPR operation, then obtain the historical value of the detection interval adjustment corresponding to the invalid value range of CPR operation as the detection interval adjustment value of the detection frame.

[0064] The interpolation resampling step size adjustment value is obtained based on the matching of the operation quality difference degree value. The specific method is as follows: obtain the preset operation quality difference degree value range and the corresponding historical value of the interpolation resampling step size adjustment in the database, and match them with the operation quality difference degree value. If the operation quality difference degree value is within a certain preset operation quality difference degree value range, then obtain the corresponding historical value of the interpolation resampling step size adjustment as the interpolation resampling step size adjustment value.

[0065] Temporal resampling interpolation is performed on the keypoint trajectory sequence corresponding to the keypoints of the operation to obtain real-time optimized keypoints and their corresponding optimized trajectory sequences. Specifically, based on the adjusted detection frame extraction interval and interpolation resampling step size, temporal resampling interpolation is performed on the keypoint trajectory sequence corresponding to the keypoints of the operation. Specifically: first, based on the adjusted detection frame extraction interval value, the original keypoint trajectory sequence is resampled, and detection frames are filtered according to the new time interval to obtain the first-stage resampled trajectory sequence; based on the adjusted interpolation resampling step size value, a unified temporal sampling axis is constructed between adjacent detection frames, and the original keypoint trajectory... The trajectory sequence is mapped onto the time axis; for the missing time points between two adjacent detection frames, the coordinate values ​​of the corresponding key points in the preceding and following detection frames are extracted, and linear interpolation is performed according to the time ratio to obtain the key point coordinates at the intermediate time, thereby generating continuous key point trajectory data; after completing all time axis interpolation, the key point coordinates corresponding to each time point are connected in chronological order to form a key point trajectory sequence, and each key point in the trajectory sequence is used as a real-time optimization key point, while its corresponding time sequence structure is used as an optimized trajectory sequence, thereby achieving the refinement and continuous enhancement of the key point trajectory in the time dimension.

[0066] By introducing the operational quality gap value as an adjustment criterion, the detection frame extraction interval adjustment value and the interpolation resampling step size adjustment value can be adaptively adjusted according to different states of CPR operation quality, thereby achieving differentiated optimization processing for different data quality states. When the operational quality gap value is small, it indicates that the current key point trajectory is relatively stable. By adjusting the detection frame extraction interval, sampling redundancy can be reduced while ensuring trajectory continuity, thus improving processing efficiency. Conversely, when the operational quality gap value is large, it indicates that the key point trajectory has significant instability or deviation. In this case, by reducing the interpolation resampling step size, more dense interpolation points can be generated between adjacent detection frames, thereby refining the key point trajectory change process and improving trajectory fitting accuracy and change capture capability. The detection frame extraction interval adjustment value can control the key point acquisition frequency at the data sampling level, reducing the computational burden in a stable state and improving the response capability in a changing state. The interpolation resampling step size adjustment value can improve the temporal resolution at the trajectory reconstruction level, making the key point trajectory smoother and more continuous, and reducing the errors caused by discrete sampling.

[0067] Furthermore, the quality value of cardiopulmonary resuscitation (CPR) operation is obtained through the following method: When the critical cause is operation occlusion, the predicted filling key point is used as the baseline key point for CPR operation; when the critical cause is trajectory tracking anomaly, the tracking optimization key point is used as the baseline key point for CPR operation. A key point motion trajectory sequence is constructed based on the baseline key points for CPR operation, and extreme point detection is performed on the key point motion trajectory sequence to identify the lowest point and rebound point during compression. The interval between two adjacent lowest points is divided into a compression cycle, resulting in a compression cycle sequence. Statistical analysis is performed on the compression cycle sequence to obtain the CPR operation execution parameter set. The CPR operation execution parameter set is compared with a preset CPR operation execution reference set to obtain comparison values. Weighted fusion processing is then performed on these comparison values ​​to obtain the CPR operation quality value.

[0068] In this embodiment, statistical analysis is performed based on the compression cycle sequence to obtain a set of CPR execution parameters. These parameters include compression depth, compression frequency, and rebound height for each compression cycle (the rebound height can be statistically determined based on the operator's hand height; sufficient rebound is only guaranteed when the operator's hand is fully released after compression). A CPR execution reference set includes reference values ​​for compression depth, compression frequency, and rebound height.

[0069] The specific method for obtaining the quality score of cardiopulmonary resuscitation (CPR) procedures is as follows:

[0070] ;

[0071] In the formula, This indicates the degree of difference in operational quality. Indicates the depth of pressure. This indicates the reference value for the pressure depth. Indicates the frequency of pressing. This indicates a reference value for the pressing frequency. Indicates the rebound height. This indicates the reference value for rebound height. Indicates the pressure depth weight. Indicates the weight of the pressing frequency. Indicates the weight of the rebound height.

[0072] The weights for compression depth, compression frequency, and rebound height can be obtained from a database. For example, historical CPR video data with standardized annotations can be collected and filtered from the database. These annotations include actual CPR operation quality scores measured manually or by equipment. Following the same method as in this scheme, the corresponding CPR operation execution parameter set, including compression depth, compression frequency, and rebound height, can be extracted from each historical video. The deviation of each parameter relative to its corresponding reference value can be calculated to construct a parameter deviation dataset. The parameter deviation data can be correlated with the corresponding actual quality scores, and a multiple linear regression method can be used for fitting training. With parameter deviation as the independent variable and quality score as the dependent variable, the regression coefficients corresponding to each parameter can be obtained. The obtained regression coefficients can be normalized so that the sum of the coefficients is 1, thereby obtaining the weights for compression depth, compression frequency, and rebound height.

[0073] By constructing a quality metric for cardiopulmonary resuscitation (CPR) operations, key parameters such as compression depth, compression frequency, and rebound height are uniformly quantified and integrated. This transforms previously scattered, multi-dimensional operational indicators into a single, comparable evaluation result, facilitating real-time assessment of CPR quality and dynamic feedback. By detecting extreme points and dividing compression cycles into key point motion trajectory sequences, the complete process of each compression can be accurately extracted. This ensures that the calculation of compression depth, frequency, and rebound height reflects a true and complete action cycle, avoiding parameter deviations caused by incomplete identification of individual compressions and improving the accuracy of parameter statistics. Depending on the critical causes, either predicted and filled key points or tracked and optimized key points are selected as baseline key points for CPR operations. This allows for targeted correction of key point data in cases of operational occlusion or trajectory tracking anomalies, effectively preventing video key point extraction errors from affecting subsequent parameter calculations. This ensures higher continuity and authenticity of the key point trajectory data used for quality assessment. Ultimately, the quality metric for CPR operations more accurately reflects actual operational skill levels, thereby improving the reliability and stability of the assessment results.

[0074] Furthermore, the improvement results of key point extraction are obtained. Specifically, the following method is used: The real-time operational key points after trajectory optimization are re-evaluated and their continuity is checked to obtain the key point extraction evaluation results for each detection frame within a preset window. If the key point extraction evaluation result is invalid, the improvement result is considered a failure. If the key point extraction evaluation result is valid, a key frame extraction parameter set is obtained, including the percentage of valid detection frames, the maximum number of consecutive valid frames, and the average confidence level. The key frame extraction parameter set is compared with a preset key frame extraction reference set. If any parameter in the key frame extraction parameter set is less than the corresponding reference set, the improvement result is considered a failure. Otherwise, an improvement degree analysis is performed based on the key frame extraction parameter set and the preset reference set to obtain a comprehensive improvement value for key point extraction. The comprehensive improvement value is compared with a preset comprehensive improvement threshold. If the comprehensive improvement value is greater than the comprehensive improvement threshold, the improvement result is considered successful; otherwise, the improvement result is considered a failure.

[0075] In this embodiment, the keyframe extraction reference set includes the effective detection frame percentage reference value, the maximum number of consecutive effective frames reference value, and the average confidence level reference value.

[0076] The key point extraction comprehensive improvement value is obtained by means of the following method:

[0077] ;

[0078] In the formula, This indicates the extraction of the comprehensive improvement value from key points. Indicates the percentage of valid detection frames. This represents the percentage of valid detected frames. Indicates the maximum number of consecutive valid frames. This represents the maximum number of consecutive valid frames. This represents the average confidence level. This represents the average confidence level control value. Indicates the weight of the proportion of valid detection frames. The weight represents the maximum number of consecutive valid frames. This represents the average confidence weight.

[0079] The effective detection frame percentage weight, maximum consecutive effective frame count weight, and average confidence weight can be obtained from a database. For example, a historical video dataset can be constructed, and the keypoint extraction results of each video before and after trajectory optimization can be processed separately. The corresponding keyframe extraction parameter sets can be calculated according to the method in this scheme, including the parameter set before optimization and the parameter set after optimization. Then, based on the changes in parameters before and after optimization, the improvement value of the effective detection frame percentage, the improvement value of the maximum consecutive effective frame count, and the improvement value of the average confidence are calculated respectively. The improvement of the three is used as an objective representation of the improvement effect of keypoint extraction, and an improvement effect dataset is constructed. The improvement value of each parameter is used as an input variable, and the overall improvement effect (using the weighted average of the three improvement values ​​as a unified evaluation index) is used as an output variable. Multiple linear regression is used for fitting training to obtain the regression coefficients corresponding to each parameter. Finally, the regression coefficients are normalized so that their sum is 1, thereby obtaining the effective detection frame percentage weight, maximum consecutive effective frame count weight, and average confidence weight respectively.

[0080] By re-analyzing the keypoint extraction evaluation results of each detection frame within a preset window, a closed-loop verification of the trajectory optimization adjustment effect is achieved. The analysis time window allows for a comprehensive evaluation of the stability and continuity of keypoint extraction over a continuous time range, avoiding interference from occasional effective or ineffective single frames on the overall judgment, thus improving the reliability of the evaluation results. When the keypoint extraction evaluation result is invalid, the improvement result is directly judged as an improvement failure because this state indicates that the keypoint data still lacks basic usability at the current stage, and further detailed analysis is meaningless. This facilitates the rapid triggering of subsequent optimization mechanisms, improving system response efficiency. Conversely, when the keypoint extraction evaluation result is effective, further comparison of the keyframe extraction parameter set with the control set and calculation of the comprehensive improvement value of keypoint extraction are used to finely distinguish between "effective but insufficient quality" cases. This avoids misjudging low-quality but barely acceptable data as successful improvement, ensuring that the data entering the subsequent cardiopulmonary resuscitation operation quality evaluation has sufficient stability and accuracy. This ensures that the keypoint extraction results meet both basic usability conditions and quality improvement requirements, thereby significantly improving the continuity and confidence level of the keypoint trajectory. By combining abnormal jump point correction, periodic trajectory reconstruction, and real-time optimization of key point generation mechanism, a complete closed loop from key point extraction, anomaly correction, trajectory optimization to effect verification is achieved. This enables the system to continuously and adaptively optimize the quality of key point data in complex video environments, thereby improving the stability, accuracy, and robustness of cardiopulmonary resuscitation operation quality assessment as a whole.

[0081] Furthermore, the execution trajectory is optimized and adjusted a second time. The specific method is as follows: Based on the comprehensive improvement value extracted from key points and the comprehensive improvement threshold, a difference is processed to obtain the comprehensive improvement difference. Based on the comprehensive improvement difference, a backtracking processing duration adjustment value is obtained. Based on this adjustment value, the first backtracking processing duration is extended to obtain the second backtracking adjustment duration. Specifically, the first backtracking processing duration is added to the adjusted backtracking processing duration to obtain the second backtracking adjustment duration. Based on the comprehensive improvement difference, a neighboring interpolation frame number adjustment value is obtained for abnormal jump points. Based on this adjustment value, the neighboring interpolation frame number is increased to obtain the second execution value of the neighboring interpolation frame number. Specifically, the current neighboring interpolation frame number is added to the adjusted neighboring interpolation frame number to obtain the second execution value of the neighboring interpolation frame number. Based on the second backtracking adjustment duration and the second execution value of the neighboring interpolation frame number, a re-analysis is performed to obtain the cardiopulmonary resuscitation operation quality value, and corresponding information reminders are issued.

[0082] In this embodiment, the difference between the comprehensive improvement value extracted from the key points and the comprehensive improvement threshold is processed to obtain the comprehensive improvement difference. Specifically, the comprehensive improvement value extracted from the key points is subtracted from the comprehensive improvement threshold to obtain the comprehensive improvement difference.

[0083] The adjustment value for the backtracking processing time is obtained based on the matching of comprehensive improvement difference. The specific method is as follows: obtain each historical comprehensive improvement difference interval stored in the database, and the historical backtracking processing time corresponding to each historical comprehensive improvement difference interval. Match the comprehensive improvement difference with each historical comprehensive improvement difference interval. If the comprehensive improvement difference is within a certain preset comprehensive improvement difference interval, then obtain the historical backtracking processing time corresponding to that comprehensive improvement difference interval as the adjustment value for the backtracking processing time.

[0084] The adjustment value of the number of neighboring interpolation frames for adjusting abnormal jump points is obtained based on the comprehensive improvement difference matching. The historical comprehensive improvement difference intervals and the historical neighboring interpolation frame adjustment values ​​corresponding to each historical comprehensive improvement difference interval are obtained from the database. The comprehensive improvement difference is matched with each historical comprehensive improvement difference interval. If the comprehensive improvement difference is within a certain preset comprehensive improvement difference interval, the historical neighboring interpolation frame adjustment value corresponding to the comprehensive improvement difference interval is obtained as the neighboring interpolation frame adjustment value for adjusting abnormal jump points.

[0085] By introducing the difference between the comprehensive improvement value and the comprehensive improvement threshold of key point extraction as the adjustment basis, adaptive control of the secondary adjustment process of trajectory optimization is achieved, thereby avoiding the problems of insufficient optimization or over-adjustment caused by using fixed parameters. Instead of directly setting a fixed secondary backtracking adjustment time, the adjustment value of the backtracking processing time is obtained based on the matching of the comprehensive improvement difference. This allows the backtracking range to dynamically change according to the degree of inadequacy of the key point extraction improvement effect: when the comprehensive improvement difference is large, it indicates that the current trajectory optimization effect is still significantly insufficient. Extending the backtracking processing time can introduce historical key point trajectory data over a longer period, thereby enhancing the stability of periodic feature detection and trend analysis, and improving the accuracy of key point prediction and correction. When the comprehensive improvement difference is small, only a small adjustment to the backtracking processing time is needed, avoiding the introduction of too much historical data leading to computational redundancy or historical noise, thus improving processing efficiency while ensuring optimization effect.

[0086] By adjusting the number of neighboring interpolation frames based on comprehensive improvement difference matching and increasing the number of neighboring interpolation frames during the correction of abnormal jump points, effective detection frame information can be obtained in a larger temporal neighborhood. This allows interpolation calculations to no longer rely solely on local adjacent frames, but rather on richer contextual data for smooth reconstruction, effectively reducing the impact of single-frame anomalies or local noise on the interpolation results and improving the smoothness and continuity of keypoint trajectory correction. When the comprehensive improvement difference is large, expanding the neighborhood interpolation range can enhance the stability of abnormal jump point correction, while maintaining a smaller neighborhood range when the difference is small avoids excessive smoothing that could lead to the loss of detailed information. Through the coordinated adaptive adjustment of the backtracking processing time and the number of neighboring interpolation frames, the secondary trajectory optimization can accurately compensate for different degrees of insufficient keypoint extraction, thereby improving the continuity, periodic consistency, and data reliability of keypoint trajectories. Ultimately, this improves the accuracy of the evaluation of CPR operation quality values ​​and enhances the robustness and adaptive optimization capabilities of the overall solution in complex video environments.

[0087] like Figure 3 As shown, Figure 3The system architecture diagram of this invention is shown below. An embodiment of this application provides a real-time cardiopulmonary resuscitation (CPR) quality assessment system based on video analysis, comprising the following modules: a key point extraction module, used for real-time video analysis and extraction of key points from CPR videos to obtain each key point, and performing effective evaluation and continuity detection based on each key point to obtain a key point extraction assessment result; an initial assessment module, used for performing CPR quality assessment to obtain a CPR operation quality value; when the key point extraction assessment result is invalid, the CPR operation quality value is marked as an invalid CPR operation value, and the key cause of the key point extraction assessment result is retrieved; otherwise, the marking is not performed; and a key cause determination module, used for performing key point prediction when the key cause is operation occlusion, to obtain a preliminary assessment result. The system performs several steps: 1) Fill in key points; 2) When the key cause is abnormal trajectory tracking, perform trajectory tracking optimization to obtain optimized key points; 3) The trajectory tracking adjustment module is used to perform real-time updates and trajectory optimization based on invalid values ​​of CPR operations, obtaining real-time optimized key points and their corresponding optimized trajectory sequences; 4) The operation evaluation and reminder module is used to obtain CPR operation quality values ​​based on predicted key point filling or tracking optimization key point analysis, and issue corresponding information reminders; 5) The secondary adjustment module is used to obtain key point extraction improvement results based on real-time optimized key points and their corresponding optimized trajectory sequences. If the key point extraction improvement result is an improvement failure, a secondary trajectory optimization adjustment is performed until the key point extraction improvement result is a successful improvement; otherwise, no processing is performed.

[0088] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A real-time assessment method for cardiopulmonary resuscitation quality based on video analytics, characterized in that: Includes the following steps: Real-time video analysis and key operation point extraction are performed on cardiopulmonary resuscitation videos to obtain each key operation point. Based on each key operation point, effective evaluation and continuity detection are carried out to obtain the key point extraction evaluation results. Perform cardiopulmonary resuscitation (CPR) quality assessment to obtain CPR operation quality values. If the key point extraction assessment result is invalid, mark the CPR operation quality value as an invalid CPR operation value and retrieve the key causes of the key point extraction assessment result; otherwise, do not perform the marking. When the critical cause is operation occlusion, perform keypoint prediction to obtain predicted fill keypoints; When the key cause is trajectory tracking anomaly, perform trajectory tracking optimization and adjustment to obtain the key points for tracking optimization. Based on the real-time update and trajectory optimization adjustment of key points for cardiopulmonary resuscitation (CPR) operation based on invalid values, the real-time optimized key points and their corresponding optimized trajectory sequences are obtained. Based on the analysis of predicted key points or tracked and optimized key points, the quality value of cardiopulmonary resuscitation operation is obtained, and corresponding information reminders are issued. Based on real-time optimization of key points and their corresponding optimization trajectory sequence analysis, the key point extraction improvement results are obtained. If the key point extraction improvement result is an improvement failure, a second adjustment of trajectory optimization is performed until the key point extraction improvement result is a successful improvement; otherwise, no processing is performed.

2. The real-time assessment method for cardiopulmonary resuscitation quality based on video analysis according to claim 1, characterized in that: The key point extraction and evaluation results are obtained through the following method: Obtain each operation key point within a preset continuous detection frame, and align key points of the same type in chronological order to obtain a temporal sequence of each operation key point; The confidence level of each operation key point in each detection frame is detected frame by frame to obtain the confidence level of each operation key point in each detection frame. The confidence level of each operation key point in each detection frame is compared with the preset confidence threshold frame by frame to obtain the validity judgment result of each detection frame. If the confidence level of each operation key point in a certain detection frame is above the confidence threshold, the validity judgment result of the detection frame is that the data is reliable; otherwise, the validity judgment result of the detection frame is that the data is unreliable. Extract the validity determination results of a preset number of consecutive detection frames. If the validity determination results of the preset number of consecutive detection frames are all reliable data, the continuity detection result is continuous detection; otherwise, the continuity detection result is interrupted detection. When the validity determination result is that the data is unreliable and / or the continuous detection result is that the detection is interrupted, the key point extraction evaluation result is that the extraction is invalid; otherwise, the key point extraction evaluation result is that the extraction is valid.

3. The real-time assessment method for cardiopulmonary resuscitation quality based on video analysis according to claim 1, characterized in that: The key factors for extracting evaluation results by retrieving key points are as follows: By tracing back from the real-time frame, the number of consecutive invalid detection frames is calculated based on the key point extraction evaluation result and recorded as the number of consecutive invalid frames. The number of consecutive invalid frames is compared with the preset occlusion judgment frame number threshold. When the number of consecutive invalid frames is above the occlusion judgment frame number threshold, the key cause of the key point extraction evaluation result is operation occlusion. When the number of consecutive invalid frames is less than the threshold for occlusion determination, the key cause of the key point extraction evaluation result is specifically trajectory tracking anomaly.

4. The real-time assessment method for cardiopulmonary resuscitation quality based on video analysis according to claim 1, characterized in that: Perform keypoint prediction to obtain predicted keypoints for filling. The specific method is as follows: The key causes of the evaluation results are retrieved from the key points, which are the detection frames corresponding to the operation occlusion. These frames are marked as operation occlusion detection frames, and the first occurrence time of the operation occlusion detection frame is recorded as the operation occlusion start time. Starting from the moment of operation occlusion, and with a preset first backtracking processing time as the backtracking execution time, data backtracking processing is performed on the key points of the operation to obtain historical key point trajectory data. The historical key point trajectory data includes the historical position sequence of each key point of the operation. Periodic feature detection is performed based on historical key point trajectory data to obtain periodic feature determination results; When the periodic feature determination result indicates the presence of periodic features, phase trend prediction is performed based on historical key point trajectory data to obtain the key point position prediction values ​​for each frame within the operation occlusion detection frame, and these values ​​are used as periodic predictions to fill in the key points. When the periodic feature determination result is that there is no periodic feature, the movement direction and movement rate of the operation key points of the detection frame are analyzed based on the historical key point trajectory data. The position of the operation key points of each detection frame in the operation occlusion detection frame is continued along the movement trend to obtain the trend prediction filling key points. Use cycle prediction to fill key points and / or trend prediction to fill key points as prediction fill key points.

5. The real-time assessment method for cardiopulmonary resuscitation quality based on video analysis according to claim 1, characterized in that: The trajectory tracking optimization and adjustment are performed to obtain key points for tracking optimization. The specific method is as follows: Smoothing is performed on the time sequence of each key operation point to obtain the smoothed key point trajectory; Anomaly detection and correction are performed on the smooth key point trajectory. Operation key points in the smooth key point trajectory whose position change amplitude exceeds the preset change amplitude threshold are identified and identified as abnormal jump points. The coordinates of the abnormal jump points are replaced with interpolated coordinates calculated based on adjacent valid detection frames to obtain the corrected operation key point motion trajectory, which is recorded as the corrected key point trajectory. Periodic feature detection is performed on the corrected keypoint trajectory to obtain the periodic feature determination result; If the periodicity feature determination result indicates the presence of periodicity features, then a preset number of reference trajectory segments are selected from the corrected key point trajectory, and periodic trajectory reconstruction is performed based on the reference trajectory segments to obtain the periodic reconstructed trajectory. If the periodicity characteristic determination result is that there is no periodicity characteristic, an early warning will be issued; The reference trajectory segment is specifically a continuous trajectory segment whose period length consistency is higher than the period length consistency threshold. The key point trajectory is corrected based on the periodic reconstructed trajectory to obtain the key point reconstructed trajectory. Each key point in the key point reconstructed trajectory is used as the tracking optimization key point, and its corresponding phase data is used as the tracking optimization key point data.

6. The real-time assessment method for cardiopulmonary resuscitation quality based on video analysis according to claim 1, characterized in that: The specific method for obtaining the real-time optimization key points and their corresponding optimization trajectory sequences is as follows: Based on the difference between the invalid value of cardiopulmonary resuscitation (CPR) operation and the preset CPR operation quality threshold, the degree of difference in operation quality is obtained. Obtain a preset threshold for the degree of difference in operation quality and compare it with the value of the degree of difference in operation quality. If the value of the degree of difference in operation quality is below the threshold for the degree of difference in operation quality, then obtain the detection interval adjustment value of the detection frame based on the invalid value matching of cardiopulmonary resuscitation operation. If the operational quality gap value is greater than the operational quality gap threshold, the interpolation resampling step size adjustment value is obtained based on the operational quality gap value. The detection frame extraction interval is shortened based on the detection interval adjustment value; Reduce the interpolation resampling step size based on the interpolation resampling step size adjustment value; Based on the adjusted detection frame extraction interval or interpolation resampling step size, time resampling interpolation processing is performed on the key point trajectory sequence corresponding to the key point of operation to obtain the real-time optimized key point and its corresponding optimized trajectory sequence.

7. The real-time assessment method for cardiopulmonary resuscitation quality based on video analysis according to claim 1, characterized in that: The specific method for obtaining the cardiopulmonary resuscitation (CPR) operation quality value is as follows: When the critical cause is operation occlusion, the predicted filling key point is used as the benchmark key point for cardiopulmonary resuscitation operation; when the critical cause is trajectory tracking anomaly, the tracking optimization key point is used as the benchmark key point for cardiopulmonary resuscitation operation. Based on the key points of the cardiopulmonary resuscitation operation benchmark, a key point motion trajectory sequence is constructed, and extreme point detection is performed on the key point motion trajectory sequence to identify the lowest point and rebound point during the compression process; Divide the interval between two adjacent lowest points into a pressing cycle to obtain a pressing cycle sequence; Based on the compression cycle sequence, statistical analysis was performed to obtain the set of cardiopulmonary resuscitation (CPR) operation execution parameters. The cardiopulmonary resuscitation (CPR) operation execution parameter set is compared with the preset CPR operation execution reference set to obtain each comparison value. Based on each comparison value, a weighted fusion process is performed to obtain the CPR operation quality value.

8. The real-time assessment method for cardiopulmonary resuscitation quality based on video analysis according to claim 1, characterized in that: The method for obtaining the improved key point extraction results is as follows: The key points of real-time operation after trajectory optimization and adjustment are re-evaluated and continuously detected to obtain the key point extraction and evaluation results of each detection frame within the preset window; When the key point extraction evaluation result is invalid, the key point extraction improvement result is improvement failure; When the key point extraction evaluation result is valid, the key frame extraction parameter set is obtained. The key frame extraction parameter set includes the percentage of valid detection frames, the maximum number of consecutive valid frames, and the average confidence level. The key frame extraction parameter set is extracted and compared with the preset key frame extraction reference set. If any parameter in the key frame extraction parameter set is less than the corresponding key frame extraction reference set, the key point extraction improvement result is an improvement failure. Otherwise, the improvement degree is analyzed based on the key frame extraction parameter set and the preset key frame extraction reference set to obtain the comprehensive improvement value of key point extraction. The improvement result is determined by comparing the comprehensive improvement value of key point extraction with the preset comprehensive improvement threshold. If the comprehensive improvement value of key point extraction is greater than the comprehensive improvement threshold, the improvement result is considered successful; otherwise, the improvement result is considered unsuccessful.

9. The real-time assessment method for cardiopulmonary resuscitation quality based on video analysis according to claim 8, characterized in that: The second adjustment of the execution trajectory optimization is specifically achieved through the following method: The comprehensive improvement value is obtained by extracting the comprehensive improvement value from key points and performing difference processing on the comprehensive improvement threshold. The backtracking processing time adjustment value is obtained based on the comprehensive improvement difference matching. The first backtracking processing time is extended based on the backtracking processing time adjustment value to obtain the second backtracking adjustment time. The neighborhood interpolation frame number adjustment value is obtained based on the comprehensive improvement difference matching. The neighborhood interpolation frame number is increased based on the neighborhood interpolation frame number adjustment value to obtain the secondary execution value of the neighborhood interpolation frame number. Based on the secondary backtracking adjustment duration and the number of neighboring interpolation frames, the secondary execution value is re-analyzed to obtain the cardiopulmonary resuscitation operation quality value, and corresponding information reminders are issued.

10. A system applying the video analysis-based real-time assessment method for cardiopulmonary resuscitation quality as described in any one of claims 1-9, characterized in that: Includes the following modules: The key point extraction module is used to perform real-time video analysis and operation key point extraction on cardiopulmonary resuscitation videos, obtain each operation key point, and perform effective evaluation and continuity detection based on each operation key point to obtain key point extraction evaluation results. The initial assessment module is used to perform cardiopulmonary resuscitation (CPR) quality assessment and obtain CPR operation quality values. When the key point extraction assessment result is invalid, the CPR operation quality value is marked as an invalid CPR operation value, and the key causes of the key point extraction assessment result are retrieved; otherwise, the marking is not performed. The critical cause determination module is used to perform key point prediction and obtain predicted filling key points when the critical cause is operation occlusion. When the key cause is trajectory tracking anomaly, perform trajectory tracking optimization and adjustment to obtain the key points for tracking optimization. The trajectory tracking and adjustment module is used to obtain and optimize the trajectory in real time based on the invalid values ​​of cardiopulmonary resuscitation operations, thereby obtaining the real-time optimized key points and their corresponding optimized trajectory sequences. The operation assessment and reminder module is used to obtain the cardiopulmonary resuscitation operation quality value based on the analysis of predicted key points or tracked and optimized key points, and to issue corresponding information reminders. The secondary adjustment module is used to obtain the key point extraction improvement result based on the analysis of the real-time optimized key points and their corresponding optimized trajectory sequences. If the key point extraction improvement result is an improvement failure, the trajectory optimization secondary adjustment is performed until the key point extraction improvement result is an improvement success; otherwise, no processing is performed.

Images